Spaces:
Sleeping
Sleeping
File size: 10,383 Bytes
ad69c81 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 | from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
import json, os, random, subprocess, tempfile, sys, time
from typing import Optional, List
from server.executor import run_code_with_tests
from server.grader import calculate_codearena_reward
app = FastAPI(title="CodeArena RL Environment")
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_methods=["*"],
allow_headers=["*"]
)
# -- State ------------------------------------------------
current_task = {}
step_count = 0
previous_attempts = []
episode_rewards = []
TASKS_DIR = os.path.join(os.path.dirname(__file__), "..", "tasks")
# -- Models -----------------------------------------------
class ResetRequest(BaseModel):
task_id: str = "easy"
buggy_code: Optional[str] = None
class StepRequest(BaseModel):
proposed_fix: str
class AgentPerformanceTracker:
def __init__(self):
self.episode_rewards = []
self.current_difficulty = "easy"
self.steps_at_difficulty = 0
def record_episode(self, avg_reward: float):
self.episode_rewards.append(round(avg_reward, 4))
if len(self.episode_rewards) > 10:
self.episode_rewards.pop(0)
self.steps_at_difficulty += 1
self._maybe_escalate()
def _maybe_escalate(self):
if self.steps_at_difficulty < 3:
return
if not self.episode_rewards:
return
avg = sum(self.episode_rewards) / len(self.episode_rewards)
old = self.current_difficulty
if avg > 0.80 and self.current_difficulty == "easy":
self.current_difficulty = "medium"
self.steps_at_difficulty = 0
elif avg > 0.75 and self.current_difficulty == "medium":
self.current_difficulty = "hard"
self.steps_at_difficulty = 0
elif avg < 0.35 and self.current_difficulty == "hard":
self.current_difficulty = "medium"
self.steps_at_difficulty = 0
elif avg < 0.35 and self.current_difficulty == "medium":
self.current_difficulty = "easy"
self.steps_at_difficulty = 0
if self.current_difficulty != old:
print(
f"[CURRICULUM] {old} -> {self.current_difficulty} "
f"after avg={avg:.3f}"
)
def get_difficulty(self) -> str:
return self.current_difficulty
def get_stats(self) -> dict:
avg = (
sum(self.episode_rewards) / len(self.episode_rewards)
if self.episode_rewards else 0.0
)
return {
"current_difficulty": self.current_difficulty,
"recent_avg_reward": round(avg, 3),
"episodes_tracked": len(self.episode_rewards),
"steps_at_current_difficulty": self.steps_at_difficulty
}
tracker = AgentPerformanceTracker()
# -- Helpers ----------------------------------------------
def load_random_task(difficulty: str):
folder = os.path.join(TASKS_DIR, difficulty)
files = [f for f in os.listdir(folder) if f.endswith(".json")]
if not files:
raise ValueError(f"No tasks found in {folder}")
path = os.path.join(folder, random.choice(files))
with open(path) as f:
return json.load(f)
def run_tests(code: str, tests: list):
passed = 0
total = len(tests)
compile_ok = True
error_log = ""
start_time = time.time()
with tempfile.NamedTemporaryFile(mode="w", suffix=".py",
delete=False, dir=tempfile.gettempdir()) as tmp:
tmp.write(code)
tmp_path = tmp.name
try:
for test in tests:
inp = test["input"]
expected = test["expected"]
# Build a test runner script
test_script = f"""
import sys
sys.path.insert(0, '')
exec(open(r'{tmp_path}').read())
inp = {repr(inp)}
if isinstance(inp, list):
result = list(locals().values())[-1](*inp) if callable(list(locals().values())[-1]) else None
# find the function
import types
funcs = {{k:v for k,v in locals().items() if isinstance(v, types.FunctionType)}}
if funcs:
fn = list(funcs.values())[0]
result = fn(*inp)
else:
result = None
else:
result = None
expected = {repr(expected)}
print('PASS' if result == expected else f'FAIL got {{result}} expected {{expected}}')
"""
runner = tempfile.NamedTemporaryFile(mode="w", suffix=".py",
delete=False, dir=tempfile.gettempdir())
runner.write(test_script)
runner.close()
try:
proc = subprocess.run(
[sys.executable, runner.name],
capture_output=True, text=True, timeout=5
)
if "PASS" in proc.stdout:
passed += 1
elif proc.returncode != 0:
compile_ok = False
error_log = proc.stderr[:300]
except subprocess.TimeoutExpired:
error_log = "Timeout"
finally:
os.unlink(runner.name)
finally:
os.unlink(tmp_path)
return compile_ok, passed, total, error_log, time.time() - start_time
def evaluate_fix(code: str, task: dict):
if "test_code" in task:
result = run_code_with_tests(
code=code,
test_code=task.get("test_code", ""),
timeout=max(float(task.get("optimal_time_seconds", 0.05)) * 10, 2.0),
)
return (
result.compile_success,
result.test_passed,
max(result.test_total, 1),
result.runtime_errors,
result.execution_time_seconds,
)
return run_tests(code, task.get("tests", []))
# -- Endpoints --------------------------------------------
@app.get("/")
def health():
return {"status": "ok", "environment": "CodeArena"}
@app.post("/reset")
def reset(req: ResetRequest):
global current_task, step_count, previous_attempts, episode_rewards
step_count = 0
previous_attempts = []
episode_rewards = []
if req.buggy_code:
# Custom mode -- user pasted their own broken code
current_task = {
"task_id": "custom",
"buggy_code": req.buggy_code,
"description": "User-provided code -- fix the bug",
"tests": [
{"input": [1, 2], "expected": None},
{"input": [0, 0], "expected": None},
{"input": [5, 5], "expected": None}
]
}
elif req.task_id in ("easy", "medium", "hard", "type_errors", "security_bugs"):
current_task = load_random_task(req.task_id)
elif req.task_id == "auto":
current_task = load_random_task(tracker.get_difficulty())
else:
current_task = load_random_task("easy")
tests_count = len(current_task.get("tests", []))
if "test_code" in current_task:
tests_count = 1
return {
"task_id": current_task["task_id"],
"curriculum_info": tracker.get_stats(),
"observation": {
"buggy_code": current_task["buggy_code"],
"error_log": "",
"test_results": f"0/{tests_count} tests passing",
"previous_attempts": []
}
}
@app.post("/step")
def step(req: StepRequest):
global step_count, previous_attempts, episode_rewards
step_count += 1
is_repeated_fix = req.proposed_fix in previous_attempts
tests = current_task.get("tests", [])
# For custom tasks with no expected values, just check compile + run
if current_task["task_id"] == "custom":
compile_ok, passed, total, error_log, execution_time_seconds = True, 0, 1, "", 0.0
try:
with tempfile.NamedTemporaryFile(mode="w", suffix=".py", delete=False) as f:
f.write(req.proposed_fix)
tmp_path = f.name
start_time = time.time()
proc = subprocess.run([sys.executable, tmp_path],
capture_output=True, text=True, timeout=5)
execution_time_seconds = time.time() - start_time
compile_ok = proc.returncode == 0
passed = 1 if compile_ok else 0
error_log = proc.stderr[:300] if not compile_ok else ""
os.unlink(tmp_path)
except Exception as e:
compile_ok = False
error_log = str(e)
else:
compile_ok, passed, total, error_log, execution_time_seconds = evaluate_fix(
req.proposed_fix,
current_task,
)
total = max(total, 1)
reward, reward_components = calculate_codearena_reward(
compile_ok=compile_ok,
passed=passed,
total=total,
execution_time_seconds=execution_time_seconds,
optimal_time_seconds=float(current_task.get("optimal_time_seconds", 0.05)),
buggy_code=current_task.get("buggy_code", ""),
proposed_fix=req.proposed_fix,
task_category=current_task.get("difficulty", current_task.get("task_id", "easy").split("-")[0]),
step_count=step_count,
is_repeated_fix=is_repeated_fix,
)
episode_rewards.append(reward)
previous_attempts.append(req.proposed_fix)
done = (reward > 0.95) or (step_count >= 5)
if done and episode_rewards:
tracker.record_episode(sum(episode_rewards) / len(episode_rewards))
tests_total = len(tests) if tests else total
return {
"reward": reward,
"done": done,
"reward_components": reward_components,
"curriculum_info": tracker.get_stats(),
"observation": {
"buggy_code": current_task["buggy_code"],
"error_log": error_log,
"test_results": f"{passed}/{tests_total} tests passing",
"previous_attempts": previous_attempts[-3:]
}
}
@app.get("/state")
def state():
return {
"task_id": current_task.get("task_id", "none"),
"step_count": step_count,
"observation": {
"buggy_code": current_task.get("buggy_code", ""),
"error_log": "",
"test_results": "",
"previous_attempts": previous_attempts
}
}
@app.get("/curriculum")
async def get_curriculum():
return tracker.get_stats()
|